The Probability in the Report: Reading NHS Genomic Test Data Without Overreading It

A genomic report can be formally correct and still be misread. It may name a gene, a transcript, a nucleotide change and a classification, yet stop short of a diagnosis. In 2026, as NHS genomic testing becomes more routine and its aggregate figures more visible, the risk is not that the data are wrong. The risk is that a laboratory finding is quietly promoted into a disease label before the clinical evidence has caught up.

The NHS Genomic Medicine Service is not a single test but a national system: samples move through Genomic Laboratory Hubs, requests are shaped by the National Genomic Test Directory, and reports return to clinicians who must decide what the result means for a patient and a family. The data it produces are process data as much as outcome data. Read them as a pathway, not a verdict.

What the service counts before it counts diagnoses

NHS genomic figures can be assembled from several points: tests requested, samples received, assays completed, reports issued, variants classified, patients discussed at a multidisciplinary team, and family cascade tests. These are not interchangeable. A count of tests is not a count of patients; a count of reports is not a count of diagnoses; a count of variants is not a count of people. A single patient may have a panel, an exome, a reanalysis and several family tests. One report may contain more than one variant.

A headline that says thousands of genomic tests were performed may be accurate and still tell you little about diagnostic yield unless the unit and denominator are stated. Genomic Medicine Service data are also shaped by eligibility. The National Genomic Test Directory sets out which patients can be offered which tests. Changes in eligibility, referral patterns and laboratory capacity can move counts without any change in disease incidence. That matters when a rising count is presented as a rising burden.

The vocabulary that decides the meaning

Much of the confusion comes from the five-part classification used in variant interpretation. A variant may be reported as pathogenic, likely pathogenic, of uncertain significance, likely benign or benign. Only the first two normally support a clinical diagnosis when they fit the phenotype. A variant of uncertain significance, or VUS, is not a weak diagnosis. It is a classification that says the evidence is insufficient to call the variant disease-causing or harmless. It may later be reclassified as evidence accumulates, but at the time of reporting it should not be acted on as if it were pathogenic.

The phrase 'a variant was found' is therefore almost empty without its classification. In a laboratory context, detection can include any change from the reference sequence. In clinical reporting, the relevant question is whether a variant explains the patient's condition. Those are different quantities. A high detection rate may reflect a broad assay and a low threshold for reporting; a lower diagnostic yield may still be clinically valuable if it identifies the few patients for whom a targeted treatment or surveillance plan exists.

Detection rate and diagnostic yield are not synonyms

Suppose 100 eligible patients are tested and a reportable variant is found in 30. That is a detection rate of 30 per cent. If only 12 of those variants are pathogenic or likely pathogenic and consistent with the clinical picture, the diagnostic yield is 12 per cent. The remaining 18 include VUS, incidental findings and variants in genes that do not fit the phenotype. A headline built on the first number would overstate the diagnostic power of the test by a factor of two and a half. It would also mislead readers about what the service has actually delivered.

Diagnostic yield is not a fixed property of a test. It depends on the patient group, the strength of the phenotype, the test used, the age of the patient and whether previous testing has already excluded common causes. A yield quoted for a highly selected cohort in a specialist clinic cannot be transferred to a broad screening population. Nor is a low yield evidence that testing is useless: in rare disease, a small number of diagnoses can justify a large number of tests, provided the pathway is honest about the odds.

From laboratory report to coded record

The journey from a laboratory report to a coded health record is where much meaning is lost or gained. The laboratory report is a clinical document. It contains the gene, the variant, the transcript, the classification, the evidence and often a recommendation for family testing or surveillance. The coded record is a summary. It may include a SNOMED CT code for a genetic test performed, a diagnosis code or a structured genomics result. It rarely carries the full nuance of the report.

This creates two opposite errors. The first is to assume that every reported variant appears as a diagnosis in routine data. It does not. The second is to assume that the absence of a diagnosis code means no diagnosis was made. It may mean the result sits in a separate genomic record, the code has not been updated, or the result is still being interpreted. Linkage between laboratory and clinical data is improving, but it is not complete. In 2026, a genomic count should be read with a data-provenance question: who recorded it, for what purpose, and at which point in the pathway?

How to read a genomic headline without turning uncertainty into disease

Start with the unit. Is the count tests, patients, families, reports or variants? Then ask for the denominator: tested patients, eligible patients or the whole population? Ask which classifications are included. A figure that includes VUS is a figure about laboratory findings, not about disease. Ask whether the number is detection or diagnostic yield, and whether the diagnosis was confirmed by clinical review. Finally, ask about time. Genomic classifications change. A VUS reported in 2021 may be reclassified in 2026; a headline that counts historical reports without updating classification will not describe current clinical meaning.

None of this makes genomic data less useful. It makes them more legible. The value of NHS genomic testing lies precisely in its ability to hold uncertainty in a structured form: to say that a variant exists, that its significance is not yet settled, and that the patient's care should proceed on clinical grounds rather than on the label alone. Read that way, a VUS is not a diagnosis waiting to happen. It is a data point with an expiry date and a duty of review. The task for readers in 2026 is not to resolve the uncertainty by headline. It is to keep the categories distinct.